This is a diagnostic we all should run. For example, take this gov spending series. Almost all of the weight is being put on positive shocks. So a LP/VAR with this series isn't really estimating the effects of general government spending, really it's spending _buildups_

So I would say your points are not unique to the new HA literature but a broader need to be cautious about linearization in general. But also reasonable to think that e.g., certainty equivalence matters more here. Hopefully deep learning will open up paths to feasible nonlinear solutions!

Another good one from Plagborg-Møller and co! #EconSky

The 1970s featured too much fiscal power ⬆️ compared to welfare-maximizing levels. Post-2008, fiscal power has been too low ⬇️ compared to first-best. Even the post-COVID inflation surge represented less-than-optimal inflation. Why? [9/12]

External validation 👍: The model's implied fiscal power closely tracks data on hours Presidents spent meeting with Fed officials 🤝 from Dreschel (2024)! Fortunately, power can be measured even after hours data end in 2008. [7/12]

This bargaining approach lets us quantify power dynamics that were previously difficult to measure. US fiscal power rose 📈 in the 60s and fell 📉 in the 80s, and spiked again 📈 after COVID. [6/12]

We can model the Fed & Treasury as playing a game, each concerned with minimizing their own policy outcomes' costs (taxes for the Treasury, inflation for the Fed) and pushing 🏛️ financing to the other. The less constrained one is, the more constrained the other must be. [3/12]

My friend and colleague Joe Anderson is on the market with an incredible #EconJMP on equilibria when government debt is issued in a way that conflicts with the central banks' objectives. One of those papers where after reading you see all relevant news through its lens! #EconSky

(🧵 below)

New working paper from Kolesár and Plagborg-Møller is a must read for anyone using VARs or LPs (so pretty much everyone in macro these days..)! #EconSky

#EconSky

Possibly biased because I am already firmly in this camp, but good new paper making the case for economists to make heavy investment into deep learning. Hope to share my experiences using PyTorch to solve highly non-linear models soon

www.galonuno.com/uploads/1/3/...